Neuron Explorer

Qwen3-0.6B

the word 'off' in phrases

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron lights up on the word 'off' in phrases like 'off the land', 'off track' or 'off chance', and goes quiet or dips on words like 'early' in time phrases. Turning it up or down did not make the model's writing about anything in particular.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    drawback of sawdust mushroom is, it’s not as nutritious for mushroom development as one might want. Accordingly, you might be frustrated with your yield on the off chance that you attempt to develop mushrooms from straight sawdust. Mushroom growing units

  2. #2 peak

    they’re competing with the farmers for land and farmers have made a lot of cattle grazing land. And as a result their natural resources such as wallabies are disappearing off the land and they’ve got to eat something. And so sheep are easy to

  3. #3 peak

    prints on a desktop printer and/or share your photos online.⏎A lot of online tutorials make it sound sort of scary, but it’s not. If you’re off a few (or 30) seconds or your water’s a couple degrees

  4. #4 peak

    in trainings and capacity development activities.<|endoftext|>One of the hardest things for a parent to do is to let their child go. And it’s not just going off to school or leaving the house after high school graduation. Once kids hit their teenage

  5. #5 peak

    the problem of hunger caused by poverty and insufficient production of food.⏎- one of the poorest countries in the world – in some of its regions, people live off less than 1.25 USD per day⏎- due to the food

  6. #6 peak

    how is it still living if it’s gone?⏎It was home to teachers, farmers, horsemen, and Kentucky civil rights leader Jesse Bottoms.⏎Located off US 60 and the Bluegrass Parkway, Huntertown was built after the

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: I have a 200,000, 200, and 20, but I didn’t have the other one. I have a 5, 5 and 5, but I haven’t had

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    Society’s 121st annual Christmas Bird Count.⏎Since 1900, an annual census of birds in the Western Hemisphere has been taken in early winter. It is conducted by volunteer birdwatchers in various areas or “count

  2. #2 peak

    survive these conditions. Watering deeply once a week, then periodically throughout can help ensure this. The best times to water your garden is early in the morning or early in the evening. You should not be watering your plants in direct sunlight.⏎If

  3. #3 peak

    It thrives in full sun but will tolerate part shade.⏎To get the highest yield of tender and flavorful leaves, pinch the tip of each branch, starting in early summer when the plants are 6" tall, to encourage bushiness. However

  4. #4 peak

    1971, six crews were identified and assigned to specific Soyuz spacecraft serial numbers for these tests. These would have been launched in three pairs beginning in early 1972, depending on the schedule for the DOS-2 station

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L17:943 suppresses
  2. L17:2228 excites
  3. L17:2969 excites
  4. L16:249 excites
  5. L4:2570 weakly suppresses
  6. L17:337 weakly suppresses
  7. L7:1440 weakly excites
  8. L16:428 weakly excites unclear: mid-sentence commas, 'most' before a verb
  9. L17:838 weakly suppresses
  10. L17:128 weakly suppresses
  11. L16:260 weakly excites
  12. L16:215 weakly suppresses
  13. L4:1776 weakly suppresses
  14. L0:1618 weakly excites
  15. L17:872 weakly excites
  16. L16:1668 weakly suppresses
  17. L3:2155 weakly excites
  18. L1:1956 weakly excites
  19. L0:2267 weakly excites
  20. L8:2842 weakly excites
  21. L10:343 weakly excites
  22. L4:911 weakly suppresses
  23. L4:2078 weakly excites
  24. L8:2567 weakly suppresses

Feeds later layers

  1. L19:879 excites
  2. L19:2848 weakly excites
  3. L22:2435 weakly excites
  4. L22:945 weakly suppresses
  5. L27:1988 weakly suppresses
  6. L21:815 weakly suppresses
  7. L21:2809 weakly excites
  8. L20:2305 weakly excites
  9. L23:2193 weakly suppresses
  10. L20:1085 weakly suppresses
  11. L25:1435 weakly excites
  12. L27:2678 weakly suppresses
  13. L21:672 weakly suppresses
  14. L23:2731 weakly excites
  15. L20:441 weakly excites
  16. L19:453 weakly suppresses
  17. L25:1115 weakly excites
  18. L24:2486 weakly excites
  19. L19:624 weakly excites
  20. L23:1661 weakly suppresses
  21. L24:197 weakly excites
  22. L23:2855 weakly excites
  23. L22:2460 weakly excites
  24. L21:1745 weakly excites

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

1 values between -14.84 and -14.281 values between -7.55 and -6.991 values between -6.99 and -6.431 values between -6.43 and -5.861 values between -5.86 and -5.303 values between -4.74 and -4.182 values between -4.18 and -3.623 values between -3.62 and -3.062 values between -3.06 and -2.505 values between -2.50 and -1.945 values between -1.94 and -1.3726 values between -1.37 and -0.81344 values between -0.81 and -0.252997 values between -0.25 and 0.31486 values between 0.31 and 0.8713 values between 0.87 and 1.434 values between 1.43 and 1.995 values between 1.99 and 2.553 values between 2.55 and 3.121 values between 4.24 and 4.802 values between 4.80 and 5.361 values between 6.48 and 7.041 values between 7.04 and 7.61offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. antar
  2. 考える
  3. insula
  4. pires
  5. ению
  6. ento
  7. ニック
  8. ees
  9. ималь
  10. ñas

Suppresses

  1. 痿
  2. withheld
  3. ynet
  4. oku
  5. cass
  6. stime
  7. getChild
  8. 人都
  9. held
  10. olph